public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Fritz Reese <fritzoreese@gmail.com>
To: fortran@gcc.gnu.org, gcc-patches@gcc.gnu.org
Subject: Re: [PATCH, Fortran] -fno-automatic with -finit-local prevents initialization of automatics in recursive functions
Date: Fri, 29 Aug 2014 20:13:00 -0000	[thread overview]
Message-ID: <CAE4aFA=ubeZ3GTWmO-tUbDqrMiUTHfy7TOsexnh01Mb9poiQ_Q@mail.gmail.com> (raw)
In-Reply-To: <CAE4aFAkSNjGYmpxUZtM8wSDYKvHaBxkpq2hX6sq8aqWfKndT+w@mail.gmail.com>

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62309

It seems with gcc-4.8.3 -fno-automatic prevents initializers from
being applied to automatic variables. The following does not behave as
I would expect it to with gcc-4.8.3 on linux-x86-64:

      function f (x)
        implicit none
        integer f, x
        integer a ! should be SAVEd from -fno-automatic
        a = a + x ! should increment by y every time
        f = a
        return
      endfunction

      recursive function g (x)
        implicit none
        integer g, x
        integer b ! should be automatic from recursive
        b = b + x ! should be set to y every time
        g = b
        return
      endfunction

      implicit none
      integer f, g

      ! Should return static value of a; accumulates x
      print *, f(3) ! -> 3, ok
      print *, f(4) ! -> 7, ok
      print *, f(2) ! -> 2, ok

      ! Should return automatic value of c; equal to y each time
      print *, g(3) ! -> garbage, expected 3
      print *, g(4) ! -> garbage, expected 4
      print *, g(2) ! -> garbage, expected 2

      end

$ gfortran -fno-automatic -finit-local-zero auto_test.f
$ ./a.out
           3
           7
           9
       32770
       32771
       32769
$

According to gfortran's manual page, -fno-automatic should "Treat each
program unit (except those marked as RECURSIVE) as if the "SAVE"
statement were specified for every local variable [...]". As far as I
can tell, -finit-local-zero should still initialize automatic
variables in RECURSIVE functions.

I believe this is a simple fix; to actually follow the specification
set forth in the man page, don't treat symbols in a RECURSIVE
namespace as if they are saved in resolve.c
(apply_default_init_local):


2014-08-29  Fritz Reese  <Reese-Fritz@zai.com>

    * resolve.c (apply_default_init_local): Don't treat variables in
    RECURSIVE units as saved.


diff --git a/gcc/fortran/resolve.c b/gcc/fortran/resolve.c
index 43eb240..a428633 100644
--- a/gcc/fortran/resolve.c
+++ b/gcc/fortran/resolve.c
@@ -11104,6 +11104,7 @@ apply_default_init_local (gfc_symbol *sym)
      result variable, which are also nonstatic.  */
   if (sym->attr.save || sym->ns->save_all
       || (gfc_option.flag_max_stack_var_size == 0 && !sym->attr.result
+          && !sym->ns->proc_name->attr.recursive
          && (!sym->attr.dimension || !is_non_constant_shape_array (sym))))
     {
       /* Don't clobber an existing initializer!  */
diff --git a/gcc/testsuite/gfortran.dg/auto_save_2.f90 b/gcc/testsuite/gfortran.
new file mode 100644
index 0000000..0d39d48
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/auto_save_2.f90
@@ -0,0 +1,52 @@
+! { dg-do run }
+! { dg-options "-fno-automatic -finit-local-zero" }
+!
+! Make sure variables are saved with -fno-automatic except in
+! functions marked RECURSIVE, and that they are still initialized with
+! -finit-local-zero.
+!
+
+function f (x)
+implicit none
+  integer f, x
+  integer a ! should be SAVEd
+  a = a + x ! should increment by y every time
+  f = a
+  return
+endfunction
+
+recursive function g (x)
+implicit none
+  integer g, x
+  integer b ! should be automatic
+  b = b + x ! should be set to y every time
+  g = b
+  return
+endfunction
+
+implicit none
+integer f, g
+
+! Should return static value of a; accumulates y
+if ( f(3) .ne. 3 ) then
+  call abort ()
+endif
+if ( f(4) .ne. 7 ) then
+  call abort ()
+endif
+if ( f(2) .ne. 9 ) then
+  call abort ()
+endif
+
+! Should return automatic value of a; equal to y each time
+if ( g(3) .ne. 3 ) then
+  call abort ()
+endif
+if ( g(4) .ne. 4 ) then
+  call abort ()
+endif
+if ( g(2) .ne. 2 ) then
+  call abort ()
+endif
+
+end


Fritz Reese

       reply	other threads:[~2014-08-29 20:13 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <CAE4aFAkSNjGYmpxUZtM8wSDYKvHaBxkpq2hX6sq8aqWfKndT+w@mail.gmail.com>
2014-08-29 20:13 ` Fritz Reese [this message]
2014-08-31 22:28   ` Tobias Burnus
2014-08-31 22:43     ` Steve Kargl

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='CAE4aFA=ubeZ3GTWmO-tUbDqrMiUTHfy7TOsexnh01Mb9poiQ_Q@mail.gmail.com' \
    --to=fritzoreese@gmail.com \
    --cc=fortran@gcc.gnu.org \
    --cc=gcc-patches@gcc.gnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).